A feedback method for smart metasurface time correlation cascaded channel

Abstract

The application relates to a feedback method for an intelligent metasurface time correlation cascade channel, comprising the following steps: S1, constructing a reconfigurable intelligent metasurface assisted multiple-input multiple-output communication system, comprising a base station, a reconfigurable intelligent metasurface and a single user terminal, determining a cascade channel model of the system, and establishing a cascade channel time-varying model; S2, constructing a channel state information compression and recovery network model based on deep learning, comprising an encoder and a decoder; after the user terminal obtains the cascade channel state information, the channel state information is compressed into code word information through the encoder, and after the base station receives the code word information, the code word is recovered into the cascade channel state information through the decoder. Compared with the prior art, the application solves the channel feedback problem of the time-varying cascade channel of the reconfigurable intelligent metasurface assisted multiple-input multiple-output system, has high information transmission accuracy, reduces the channel feedback overhead, has wide application scenarios, good applicability, lower complexity and better network performance.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a feedback method for intelligent metasurface time-correlation cascaded channels. Background Technology

[0002] In wireless communication, the ability of the base station to obtain correct channel state information directly affects the communication quality of the system. Since smart metasurfaces only reflect signals and lack signal processing capabilities, obtaining channel state information for cascaded channels is quite difficult. Furthermore, the sheer number of components within the smart metasurface itself exacerbates this challenge.

[0003] In existing technologies, methods for base stations to obtain channel state information (CSO) of intelligent metasurface cascaded channels mainly fall into two categories: channel estimation and channel feedback. Channel estimation is primarily used in time-division multiplexing systems, where the uplink channel can be estimated at the base station to obtain downlink CSO information due to channel reciprocity. However, in frequency-division multiplexing systems, since channel reciprocity does not hold, users need to obtain downlink CSO information and then provide feedback. This feedback process significantly occupies the channel and impacts communication quality. Therefore, it is necessary to compress the channel state information to reduce feedback overhead.

[0004] Current feedback methods do not meet the requirements of low complexity and high accuracy, and do not take into account the time correlation of cascaded channels. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a low-complexity and high-accuracy feedback method for intelligent metasurface time-correlation cascaded channels.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] This invention provides a feedback method for time-correlation cascaded channels of intelligent metasurfaces, the method comprising the following steps:

[0008] Step S1: Construct a reconfigurable smart metasurface-assisted multiple-input multiple-output communication system, including a base station, a reconfigurable smart metasurface, and a single user terminal; determine the cascaded channel model of the system; and establish a time-varying model of the cascaded channel.

[0009] Step S2: Construct a deep learning-based channel state information compression and recovery network model, including an encoder and a decoder; after the user terminal obtains the cascaded channel state information, it compresses the channel state information into codeword information through the encoder, and after the base station receives the codeword information, it uses the decoder to recover the codeword into cascaded channel state information.

[0010] Preferably, the reconfigurable smart metasurface-assisted large-scale multiple-input multiple-output communication system in step S1 includes a base station, a reconfigurable smart metasurface, and a single user terminal; the base station includes multiple transmitting antennas and multiple receiving antennas, the smart metasurface includes multiple reflecting elements, and the direct link between the base station and the user terminal is blocked by an obstacle.

[0011] Preferably, the mathematical expression of the cascaded channel model of the system in step S1 is:

[0012] G = diag(h) H )H

[0013] In the formula, G represents the cascaded channel, H represents the channel state information from the smart metasurface to the base station, and h H This represents the channel state information from the user terminal to the smart metasurface;

[0014] The cascaded channel time-varying model:

[0015] G t =(1-α) 2 )G t-1 +α 2 U(t)

[0016] Where α∈[0,1) represents the correlation coefficient, when α→1 it means the cascaded channel has no correlation, when α→0 it means the cascaded channel is a time-invariant channel; G t G t-1 Let U(t) be the cascaded channel at time t and time t-1, respectively, and let U(t) be the noise.

[0017] Preferably, the expression for the received signal y of the base station is:

[0018]

[0019]

[0020]

[0021] In the formula, h represents the channel state information from the user terminal to the smart metasurface, e is the phase shift matrix of the smart metasurface, H represents the channel state information from the smart metasurface to the base station, p represents the user's transmit power, v is the user's precoding vector, x is the information transmitted by the user, and n is the additive white Gaussian noise at the user; the superscript H indicates the conjugate transpose operation; ρ i and ξ i Let m(p) represent the path gain of the i-th path, and L1 and L2 represent the number of paths from the user to the smart metasurface and from the smart metasurface to the base station, respectively; 1,i ,q 1,i ), Let represent the transfer vector from the smart metasurface to the base station, the antenna array response of the i-th path, and the transfer vector from the user to the smart metasurface, respectively.

[0022] Preferably, the transfer vector m(p) from the smart metasurface to the base station 1,i ,q 1,i Antenna array response for the i-th path And the transfer vector m from the user to the smart metasurface H (p 2,i ,q 2,i The expressions are as follows:

[0023]

[0024]

[0025]

[0026] In the formula, n1∈{1,2,…,N1}, n2∈{1,2,…,N2}, m1∈{1,2,…,M}, N1 and N2 represent the number of elements in the smart metasurface in the horizontal and vertical directions, respectively, and M is the number of base station transmission antennas; and These are the normalized spatial azimuth and elevation angles of the intelligent metasurface, ranging from... Inside, α BR,i β BR,i These are the azimuth and elevation angles of the signal transmitted from the base station to the RIS, respectively. It is the normalized spatial azimuth of the base station, ranging from Inside, λ is the origin angle of the signal transmitted from the base station to the RIS; λ is the wavelength; d1 and d2 represent the distance between the smart metasurface elements and the antenna spacing at the base station, respectively. and These are the normalized spatial azimuth and elevation angles of the intelligent metasurface, respectively, and their ranges are all within... Inner; α RU,i β RU,i These are the azimuth and elevation angles of the signal originating from the RIS and being transmitted to the user.

[0027] Preferably, the encoder in step S2 includes a long short-term memory convolutional layer, a first reshaping module, and a feature compression module; wherein, the feature compression module includes a parallel long short-term memory module and a fully connected layer; after the user terminal obtains the cascaded channel state information, it is convolved by the long short-term memory convolutional module and converted into a matrix form, the matrix is ​​converted into a vector by the first reshaping module, and then the vector is compressed into a codeword of a set length by the feature compression module.

[0028] Preferably, the decoder includes a feature decompression module, a second reshaping module, and a refinement network connected in sequence; the feature decompression module includes a parallel long short-term memory module and a fully connected layer; after the base station receives the codeword information, it first restores it to a vector through the feature decompression module, and then converts it into a matrix by the second reshaping module. After restoring it to matrix form, the restored result needs to be input into the refinement network for feature restoration; wherein, the refinement network includes an attention convolution module, a batch normalization module, and a fully connected layer with Leaky ReLU activation function.

[0029] Preferably, the attention convolution module includes an attention mechanism module and a separate convolution module;

[0030] The attention mechanism module includes a global average pooling layer and two fully connected layers connected in sequence. The two fully connected layers are activated by the ReLU function and the Sigmoid function in sequence. The final result is multiplied by the feature map generated by the separable convolution module.

[0031] The separate convolution module includes a depthwise convolution module and a pointwise convolution module. After receiving the feature map processed by the depthwise convolution, the pointwise convolution module processes the information at the same position on the feature map and generates a new feature map.

[0032] According to a second aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement any of the methods described above.

[0033] According to a third aspect of the invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements any of the methods described above.

[0034] Compared with the prior art, the present invention has the following advantages:

[0035] 1) This invention treats the cascaded channel as an image and uses deep learning to compress and restore the channel. Deep learning has powerful learning capabilities, excellent processing capabilities in complex scenarios, and good portability and scalability. Therefore, this invention can adapt to complex scenarios and has good robustness.

[0036] 2) This invention introduces an attention mechanism to enhance the extraction of information from the cascaded channel and reduces the network model and network complexity by using depthwise separable convolution. Therefore, this invention has low complexity and high accuracy, which can effectively reduce the overhead of cascaded channel feedback and ensure the correctness of the channel state information of the cascaded channel obtained by the base station.

[0037] 3) This invention introduces a long short-term memory algorithm, which can utilize the time correlation of time-varying channels to assist in the compression and recovery process of cascaded channels. Therefore, this invention has a fast convergence speed. Attached Figure Description

[0038] Figure 1 This is a flowchart of the method of the present invention;

[0039] Figure 2 Diagram of a large-scale multiple-input multiple-output system assisted by intelligent metasurfaces;

[0040] Figure 3 This is a network structure diagram of the encoder;

[0041] Figure 4 This is a network structure diagram of the decoder;

[0042] Figure 5 Here is a diagram of the internal structure of the Attention-conv module;

[0043] Figure 6 This represents the normalized mean square error of the recovery results at different compression ratios under the same correlation coefficient.

[0044] Figure 7 This represents the normalized mean square error of the recovery results at the same compression ratio under different correlation coefficients. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0046] Example

[0047] like Figure 1 As shown in the figure, this embodiment presents a feedback method for time-dependent cascaded channels of smart metasurfaces, applicable to large-scale multiple-input multiple-output systems assisted by smart metasurfaces, used for feedback of time-varying smart metasurface cascaded channels. The method includes the following steps:

[0048] Step S1: Construct a reconfigurable smart metasurface-assisted multiple-input multiple-output communication system, including a base station (BS), a reconfigurable smart metasurface (RIS), and a single user terminal (UE). Determine the cascaded channel model of the system and establish a time-varying model of the cascaded channel.

[0049] Step S2: Construct a deep learning-based channel state information compression and recovery network model, including an encoder and a decoder; after the user terminal obtains the cascaded channel state information, it compresses the channel state information into codeword information through the encoder, and after the base station receives the codeword information, it uses the decoder to recover the codeword into cascaded channel state information.

[0050] The method of the present invention will now be described in detail.

[0051] like Figure 2 As shown, the reconfigurable smart metasurface-assisted large-scale multiple-input multiple-output communication system in step S1 includes a base station, a reconfigurable smart metasurface, and a single user terminal. The base station includes multiple transmit antennas and multiple receive antennas, the smart metasurface includes multiple reflective elements, and the direct link between the base station and the user terminal is blocked by an obstacle. In this embodiment, the base station has M transmit antennas, the smart metasurface has N reflective elements, and the user terminal of the transmission system is a single user.

[0052] The cascaded channel state of this invention is time-varying. Because the user's movement speed is limited, the user's movement distance is limited within a short time (e.g., within the feedback interval), and the channel change is also limited. Assuming the user's movement speed is 360 kilometers per hour, the movement distance within one millisecond is only 0.1 meters. Therefore, the user's surrounding environment will not completely change. Since the channel state is determined by the propagation environment, the channel state information of adjacent time slots has a high correlation. Assuming the user can obtain perfect channel state information, in a time-varying scenario, a cascaded channel model and a time-varying cascaded channel model are established for the system:

[0053] The mathematical expression for the cascaded channel model of the system is:

[0054] G = diag(h) H )H

[0055] In the formula, G represents the cascaded channel, H represents the channel state information from the smart metasurface to the base station, and h H This represents the channel state information from the user terminal to the smart metasurface;

[0056] Cascaded channel time-varying model:

[0057] G t =(1-α) 2 )G t-1 +α 2 U(t)

[0058] Where α∈[0,1) represents the correlation coefficient, when α→1 it means the cascaded channel has no correlation, when α→0 it means the cascaded channel is a time-invariant channel; G t G t-1Let U(t) be the cascaded channel at time t and time t-1, respectively, and let U(t) be the noise.

[0059] The expression for the received signal y at the base station is:

[0060]

[0061]

[0062]

[0063] In the formula, h represents the channel state information from the user terminal to the smart metasurface, e is the phase shift matrix of the smart metasurface, H represents the channel state information from the smart metasurface to the base station, p represents the user's transmit power, v is the user's precoding vector, x is the information transmitted by the user, and n is the additive white Gaussian noise at the user; h is an N-row column vector, H is an N-row M-column matrix, e is an N-column row vector, and v is an M-row column vector; the superscript H indicates the conjugate transpose operation; ρ i and ξ i Let m(p) represent the path gain of the i-th path, and L1 and L2 represent the number of paths from the user to the smart metasurface and from the smart metasurface to the base station, respectively; 1, ,q 1, ), Let represent the transfer vector from the smart metasurface to the base station, the antenna array response of the i-th path, and the transfer vector from the user to the smart metasurface, respectively, with the following expressions:

[0064]

[0065]

[0066]

[0067] In the formula, n1∈{1,2,…,N1}, n2∈{1,2,…,N2}, m1∈{1,2,…,M}, N1 and N2 represent the number of elements in the smart metasurface in the horizontal and vertical directions, respectively, and M is the number of base station transmission antennas; and These are the normalized spatial azimuth and elevation angles of the intelligent metasurface, ranging from... Inside, α BR, β BR, These are the azimuth and elevation angles of the signal transmitted from the base station to the RIS, respectively. It is the normalized spatial azimuth of the base station, ranging from Inside, λ is the origin angle of the signal transmitted from the base station to the RIS; λ is the wavelength; d1 and d2 represent the distance between the smart metasurface elements and the antenna spacing at the base station, respectively. and These are the normalized spatial azimuth and elevation angles of the intelligent metasurface, respectively, and their ranges are all within... Inner; α RU, β RU, These are the azimuth and elevation angles of the signal originating from the RIS and being transmitted to the user.

[0068] This invention aims to improve the accuracy of channel state information (CSO) recovery after compression. It treats the cascaded channel as an image and constructs a neural network with an encoder and decoder. The input to the neural network is the CSO matrix of the cascaded channel. In this example, it is assumed that N = M = 32, and the CSO is divided into real and imaginary parts. The CSO information from four adjacent time slots is grouped together, resulting in an input matrix shape of 4 × 32 × 32 × 2. After the user obtains the cascaded CSO information, the encoder compresses it. Upon receiving the compressed codewords from the user, the base station uses a decoder to recover the cascaded CSO information from the codewords.

[0069] In the encoder, long short-term memory (LSTM) convolution is used to extract channel information. LSTM convolution is a variant of the LSM algorithm used to address the vanishing gradient problem of time series gradients as computation time increases. It transforms the calculation of weights from a linear operation to a convolution operation, inheriting the capabilities of the LSM algorithm to capture the temporal correlation of the channel while also describing and extracting local details from the channel features.

[0070] like Figure 3 As shown, the encoder includes a long short-term memory convolutional layer, a first reshaping module, and a feature compression module; wherein, the feature compression module includes a parallel long short-term memory module and a fully connected layer; after the user terminal obtains the cascaded channel state information, it is convolved by the long short-term memory convolutional module and converted into a matrix form, the matrix is ​​converted into a vector of length 2048 by the first reshaping module, and then the vector is compressed into a codeword of length M by the feature compression module.

[0071] like Figure 4 As shown, the decoder includes a feature decompression module, a second reshaping module, and a refinement network connected in sequence. The feature decompression module includes a parallel long short-term memory module and a fully connected layer. After the base station receives the codeword information, it first uses the feature decompression module to recover it into a vector, and then the second reshaping module converts it into a matrix. After recovering it into matrix form, the recovered result needs to be input into the refinement network for feature recovery. The refinement network includes an attention convolution module, a batch normalization module, and a fully connected layer with Leaky ReLU activation function. It is worth noting that in this embodiment, the recovered result needs to be processed by the refinement network twice for feature recovery.

[0072] like Figure 5 As shown, the attention-based convolution module includes an attention mechanism module and a separable convolution module. The attention mechanism originates from research on human vision. In cognitive science, due to information processing bottlenecks, humans selectively focus on some information while ignoring others. In channel state information, the density and importance of information vary; some information directly affects the recovery result, while the impact of other information is negligible. Therefore, an attention mechanism can be introduced to enhance the processing of some information while ignoring unimportant information, improving network performance while reducing information processing time.

[0073] Figure 6 and Figure 7 These are the normalized mean square errors of the recovery results at different compression ratios under the same correlation coefficient, and the normalized mean square errors of the recovery results at the same compression ratio under different correlation coefficients, respectively.

[0074] The attention mechanism module consists of a globally average pooling layer and two fully connected layers connected sequentially. The two fully connected layers are activated by the ReLU function and the Sigmoid function, respectively. The final result is multiplied by the feature map generated by the separable convolution. The separable convolution module comprises two steps: depthwise convolution and pointwise convolution. The depthwise convolution is a set of convolutions. When M feature maps are received, the depthwise convolution uses M 3x3x3 3D ​​convolutions to process these M feature maps separately. The pointwise convolution module is a 1x1x1 3D convolution of depth M. After receiving the feature maps processed by the depthwise convolution, the pointwise convolution processes the information at the same location in these M maps and generates N new feature maps. Compared to directly using 3D convolution, separable convolution improves network performance while reducing network parameters and complexity.

[0075] The system described in this example receives signals under Ricean channel and perfect channel state information conditions, and no errors occur during the process of feeding back cascaded channel state information at the user end.

[0076] The electronic device of this invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0077] Multiple components in the device are connected to the I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0078] The processing unit executes the various methods and processes described above, such as methods S1-S2. For example, in some embodiments, methods S1-S2 may be implemented as computer software programs tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of methods S1-S2 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1-S2 by any other suitable means (e.g., by means of firmware).

[0079] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload programmable logic devices (CPLDs), and so on.

[0080] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0081] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0082] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A feedback method for time-correlation cascaded channels of intelligent metasurfaces, characterized in that, The method includes the following steps: Step S1: Construct a reconfigurable smart metasurface-assisted multiple-input multiple-output communication system, including a base station, a reconfigurable smart metasurface, and a single user terminal; determine the cascaded channel model of the system; and establish a time-varying model of the cascaded channel. Step S2: Construct a deep learning-based channel state information compression and recovery network model, including an encoder and a decoder; after the user terminal obtains the cascaded channel state information, it compresses the channel state information into codeword information through the encoder, and after the base station receives the codeword information, it uses the decoder to recover the codeword into cascaded channel state information. The mathematical expression for the cascaded channel model of the system in step S1 is: In the formula, For cascaded channels, This represents the channel state information from the smart metasurface to the base station. This represents the channel state information from the user terminal to the smart metasurface; The cascaded channel time-varying model: in, Represents the correlation coefficient, when When it means that the cascaded channels are completely uncorrelated, when The time indicates that the cascaded channel is a time-invariant channel; They are respectively time, Cascaded channels at any time For noise; The base station receives signals The expression is: In the formula, The phase shift matrix of the intelligent metasurface. This represents the channel state information from the smart metasurface to the base station. Indicates the user's transmit power. It is the user's precoded vector. For information sent by users, It is additive white Gaussian noise at the user's location; the superscript * indicates the conjugate transpose operation; and Indicates the first Path gain of the path, and These represent the number of paths from the user to the smart metasurface and from the smart metasurface to the base station, respectively. , , Represent the transfer vector from the smart metasurface to the base station, respectively. The antenna array response along the path and the transfer vector from the user to the smart metasurface; The transfer vector from the smart metasurface to the base station , No. Antenna array response along a single path and the transfer vector from the user to the smart metasurface The expressions are as follows: In the formula, , , , and These represent the number of components in the horizontal and vertical directions of the smart metasurface, respectively. The number of transmission antennas for the base station; and These are the normalized spatial azimuth and elevation angles of the intelligent metasurface, ranging from... Inside, These are the azimuth and elevation angles of the signal transmitted from the base station to the RIS, respectively. It is the normalized spatial azimuth of the base station, ranging from Inside, The origin angle of the signal transmitted from the base station to the RIS; It's the wavelength. and These represent the distances between smart metasurface components and the antenna spacing at the base station, respectively. and These are the normalized spatial azimuth and elevation angles of the intelligent metasurface, respectively, and their ranges are all within... Inside; , These are the azimuth and elevation angles of the signal originating from the RIS and being transmitted to the user.

2. The feedback method for a time-correlation cascaded channel for intelligent metasurfaces according to claim 1, characterized in that, The reconfigurable smart metasurface-assisted large-scale multiple-input multiple-output communication system in step S1 includes a base station, a reconfigurable smart metasurface, and a single user terminal; the base station includes multiple transmit antennas and multiple receive antennas, the smart metasurface includes multiple reflective elements, and the direct link between the base station and the user terminal is blocked by an obstacle.

3. The feedback method for a time-correlation cascaded channel for intelligent metasurfaces according to claim 1, characterized in that, The encoder in step S2 includes a long short-term memory convolutional layer, a first reshaping module, and a feature compression module. The feature compression module includes a parallel long short-term memory module and a fully connected layer. After the user terminal obtains the cascaded channel state information, it is convolved by the long short-term memory convolution module and converted into a matrix form. The matrix is ​​then converted into a vector by the first reshaping module, and finally compressed into a codeword of a set length by the feature compression module.

4. The feedback method for a time-correlation cascaded channel for intelligent metasurfaces according to claim 3, characterized in that, The decoder includes a feature decompression module, a second reshaping module, and a refinement network connected in sequence; the feature decompression module includes a parallel long short-term memory module and a fully connected layer. After receiving the codeword information, the base station first restores it to a vector through the feature decompression module, and then converts it into a matrix through the second reshaping module. After being restored to matrix form, the restored result needs to be input into the refinement network for feature restoration. The refinement network includes an attention convolution module, a batch normalization module, and a fully connected layer with Leaky ReLU activation function.

5. The feedback method for a time-correlation cascaded channel for intelligent metasurfaces according to claim 4, characterized in that, The attention convolution module includes an attention mechanism module and a separate convolution module; The attention mechanism module includes a global average pooling layer and two fully connected layers connected in sequence. The two fully connected layers are activated by the ReLU function and the Sigmoid function in sequence. The final result is multiplied by the feature map generated by the separable convolution module. The separate convolution module includes a depthwise convolution module and a pointwise convolution module. After receiving the feature map processed by the depthwise convolution, the pointwise convolution module processes the information at the same position on the feature map and generates a new feature map.

6. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 5.

Images